Discovering Causal Rules in Knowledge Graphs using Graph Embeddings

Discovering causal relationships is the goal of many experiments in science. To discover these relationships in observational data, the potential outcome framework is widely used. Within this framework, a recent approach uses Knowledge Graphs (KGs) to discover causal rules embedded within observatio...

Full description

Saved in:
Bibliographic Details
Published in2022 IEEE/WIC/ACM International Joint Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT) pp. 95 - 102
Main Authors Simonne, Lucas, Pernelle, Nathalie, Sais, Fatiha, Thomopoulos, Rallou
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.11.2022
Subjects
Causal Rules
Explainability
Graph Embeddings
Hybrid AI
Intelligent agents
Knowledge Discovery
Knowledge engineering
Knowledge graphs
Ontologies
Semantics
Task analysis
Online AccessGet full text

Cover

More Information
Summary:Discovering causal relationships is the goal of many experiments in science. To discover these relationships in observational data, the potential outcome framework is widely used. Within this framework, a recent approach uses Knowledge Graphs (KGs) to discover causal rules embedded within observational data. Such rules were found to express the following relationship: that differences in treatments lead to differences in a studied characteristic. However, this framework relies on matching similar instances which is challenging in knowledge graphs since instance descriptions can be complex, incomplete, or erroneous. To discover differential causal rules in KGs, this paper presents a new approach that uses a matching method based on graph embeddings. Our experiments on KGs from two different domains show that our approach is robust to incomplete KGs. Not only that, but compared to the state of the art, it is able to discover for a studied characteristic meaningful rules that explain many more differences than previously found.
DOI:10.1109/WI-IAT55865.2022.00023